Conference Proceedings

Seven Numeric Properties of Effectiveness Metrics

A Moffat

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Springer Verlag | Published : 2013

Abstract

Search effectiveness metrics quantify the relevance of the ranked document lists returned by retrieval systems. In this paper we characterize metrics according to seven numeric properties - boundedness, monotonicity, convergence, top-weightedness, localization, completeness, and realizability. We demonstrate that these properties partition the commonly-used evaluation metrics, and hence provide a framework in which the relationships between effectiveness metrics can be better understood, including their relative merits for different applications. © 2013 Springer-Verlag.

University of Melbourne Researchers